[Enhancement] Create a Claude skill from the InSituPy MCP server

[jwrth]

Summary

The InSituPy MCP server currently provides Claude with dynamic, on-demand access to API documentation, source code, and workflows at runtime. A complementary Claude skill (a SKILL.md file) would make this knowledge available upfront - without requiring tool calls - improving response quality and reducing latency for common coding tasks.

Motivation

• The MCP server requires a running process and tool calls for every lookup
• A skill bakes in the most important patterns so Claude "knows" them before writing any code
• Skills and the MCP server are complementary: skill for common patterns, MCP for deep/dynamic introspection
• Reduces the need to re-explain InSituPy conventions in every chat session

Proposed Strategy

Phase 1 - Content extraction

Use the MCP server's own tools to generate the skill content:

get_workflow_guide       → Standard workflows section
get_public_api           → API overview and submodule structure
get_plotting_api         → Plotting conventions
get_preprocessing_api    → Preprocessing conventions
get_tools_api            → Analysis tools
get_data_model           → InSituData/InSituExperiment structure
get_io_formats           → Supported input formats
get_result_types         → Output object patterns
get_spatialdata_api      → SpatialData interoperability
get_interactive_guide    → napari viewer usage
get_datasets_guide       → Sample datasets for examples

Phase 2 - Distillation

Condense the raw output into a concise SKILL.md with these sections:

1. When to use this skill - trigger description for Claude's routing
2. Core data model - InSituData, InSituExperiment, key attributes (cells.table, images, annotations)
3. Common workflows - read → preprocess → cluster → visualize, multi-sample DGE, image registration
4. API conventions - submodule shorthands (ispy.pl, ispy.pp, ispy.tl, ispy.io, ispy.im), parameter naming patterns
5. Key gotchas - lazy loading behavior, pixel size requirements, segmentation layer switching, AnnData access patterns
6. SpatialData interoperability - conversion patterns
7. Code examples - one representative snippet per major workflow

Phase 3 - Integration and testing

• Place SKILL.md in /mnt/skills/user/insitupy/ (or repo equivalent)
• Test by prompting Claude with typical InSituPy tasks and verifying it applies patterns correctly without needing MCP tool calls
• Iterate: add gotchas and examples discovered during testing

Notes

• The skill should be kept concise - ideally under ~500 lines - to avoid bloating Claude's context
• The MCP server remains valuable for deep lookups (source code, full docstrings, test files) that don't belong in a skill
• The skill can be regenerated periodically as the API evolves, using the same MCP-based extraction approach